zhangyiss/mlx-examples
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx-examples.git synced 2025-06-25 01:41:19 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

updated results (#165)

Browse Source
This commit is contained in:
Sarthak Yadav 2023-12-21 15:30:17 +01:00 committed by GitHub
parent 22620de3ee
commit 4addd02988
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 11 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
speechcommands/README.md
View File

@ -40,22 +40,22 @@ python main.py --help
## Results ## Results
After training with the `kwt1` architecture for 10 epochs, you After training with the `kwt1` architecture for 100 epochs, you
should see the following results: should see the following results:
``` ```
Epoch: 9 | avg. Train loss 0.519 | avg. Train acc 0.857 | Throughput: 661.28 samples/sec Epoch: 99 | avg. Train loss 0.018 | avg. Train acc 0.996 | Throughput: 662.51 samples/sec
Epoch: 9 | Val acc 0.861 | Throughput: 2976.54 samples/sec Epoch: 99 | Val acc 0.893 | Throughput: 3091.26 samples/sec
Testing best model from epoch 9 Testing best model from epoch 97
Test acc -> 0.841 Test acc -> 0.882
``` ```
For the `kwt2` model, you should see: For the `kwt2` model, you should see:
``` ```
Epoch: 9 | avg. Train loss 0.374 | avg. Train acc 0.895 | Throughput: 395.26 samples/sec Epoch: 99 | avg. Train loss 0.003 | avg. Train acc 1.000 | Throughput: 396.53 samples/sec
Epoch: 9 | Val acc 0.879 | Throughput: 1542.44 samples/sec Epoch: 99 | Val acc 0.901 | Throughput: 1543.48 samples/sec
Testing best model from epoch 9 Testing best model from epoch 94
Test acc -> 0.861 Test acc -> 0.893
``` ```
Note that this was run on an M1 Macbook Pro with 16GB RAM. Note that this was run on an M1 Macbook Pro with 16GB RAM.
@ -65,5 +65,5 @@ schedules, which is used along with the AdamW optimizer in the official
implementation. We intend to update this example once these features are added, implementation. We intend to update this example once these features are added,
as well as with appropriate data augmentations. as well as with appropriate data augmentations.
[^1]: Based one the paper [Keyword Transformer: A Self-Attention Model for Keyword Spotting](https://www.isca-speech.org/archive/interspeech_2021/berg21_interspeech.html) [^1]: Based on the paper [Keyword Transformer: A Self-Attention Model for Keyword Spotting](https://www.isca-speech.org/archive/interspeech_2021/berg21_interspeech.html)
[^2]: We use version 0.02. See the [paper]((https://arxiv.org/abs/1804.03209) for more details. [^2]: We use version 0.02. See the [paper](https://arxiv.org/abs/1804.03209) for more details.